Neuron Explorer

Qwen3-0.6B

comma closing an introductory phrase

A neuron in layer 11 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron lights up on a comma that ends a short opening phrase, like 'Through elective courses,' or 'By getting the help of a tutor,', right before the main sentence starts. When it was turned up, the model's writing did not show a clear theme.

Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).

Fires most on

The text windows where this neuron's activation was highest. Amber marks positive activation; the outlined token is the peak. Hover a token for its strength.

  1. #1 peak

    sub-matrix format.⏎The data file format for nmrPipe files is the serial format. It is possible, by creating an appropriate parameter file, for NMRView to read these files directly. It is, however, preferable

  2. #2 peak

    guide and recommended materials as a resource to respond to children’s interest and curiosity as they explore the properties of everyday substances found in the kitchen. Through planned explorations, teachers will engage children in the process of scientific inquiry. Teacher background information, specific

  3. #3 peak

    to protect endangered species and environments. Specifically, our program includes courses in forestry, soils, ecology, fisheries and wildlife management, biology and plants. Through elective courses, you can focus your education on agronomy, animal science or wildlife biology.⏎This

  4. #4 peak

    be overly concerned about long-term damage to joints such as knees, ankles and hips, as a result of running because with good quality running shoes and a sensible approach, the risks are minimal, and the benefits of being fitter outweigh those possible

  5. #5 peak

    time for students. The student can register for a few hours with a specific tutor or they can just stop by for help. By getting the help of a tutor, the student can better understand their coursework.⏎Check the Textbook⏎Many textbooks will

  6. #6 peak

    Empire. The final conquest of the Byzantines was not achieved until 1453 with the fall of Constantinople (now Istanbul), but by that date all the surrounding territory was in Ottoman hands.⏎The initial areas of expansion under O

Steering

Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.

turned down a lot
Prompt

Here is something I have been thinking about lately: I am going to say that the time to do this, I am going to say that I am going to say that I am going to say that I am going to say that I am going to say that I am going to say that

Most negative on

Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.

  1. #1 peak

    tools.⏎IF continues to produce a steady stream of strongly RPG-influenced works. Quite apart from the continuing popularity of old<|endoftext|>Please Note: By clicking a link to any resource listed

  2. #2 peak

    help of an overnight sleep study in New Jersey at Sleep Dynamics. Schedule a consultation with a sleep medicine specialist today by calling (848) 217-

  3. #3 peak

    scientists' understanding of ATP's role in taste. "It's certainly an important piece of the puzzle," neuroscientist Sue Kinnamon of the University of Colorado, Denver<|endoftext|>Videos > Rule 81: Causal Clauses⏎Causal

  4. #4 peak

    "crook."⏎Partly this is the Denial Paradox. When a guy denies he's a crook, what goes through most people's minds is, "why does he have to deny it? There must be some

Connections

Show on map

Which neurons in earlier layers feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.

Fed by earlier layers

  1. L8:1377 excites
  2. L10:371 suppresses
  3. L10:2089 excites
  4. L7:884 excites
  5. L10:422 excites
  6. L10:747 suppresses
  7. L7:2984 suppresses
  8. L8:214 weakly excites
  9. L7:1134 weakly suppresses
  10. L7:1833 weakly suppresses
  11. L3:565 weakly excites
  12. L7:96 weakly excites
  13. L7:1924 weakly suppresses
  14. L10:3030 weakly excites
  15. L7:1141 weakly suppresses
  16. L9:943 weakly suppresses passive 'formed by' and 'based on' phrasing
  17. L10:501 weakly suppresses
  18. L9:1002 weakly excites comma after who, when or where
  19. L10:720 weakly excites
  20. L8:359 weakly suppresses
  21. L8:2433 weakly excites
  22. L8:511 weakly excites
  23. L6:234 weakly excites
  24. L7:1301 weakly excites

Feeds later layers

  1. L12:1642 excites
  2. L15:1543 excites
  3. L14:2944 excites
  4. L12:2685 excites
  5. L16:1531 excites
  6. L14:1547 excites
  7. L12:1137 suppresses
  8. L13:1355 excites
  9. L19:2335 excites
  10. L15:1943 suppresses
  11. L13:1375 excites
  12. L19:2877 excites
  13. L14:620 suppresses
  14. L14:1158 excites 'to' starting a purpose clause
  15. L16:1332 excites
  16. L19:544 suppresses
  17. L12:2704 excites
  18. L20:246 suppresses
  19. L21:1556 excites
  20. L12:2212 excites 'to' or 'for' opening a purpose phrase
  21. L14:1887 suppresses
  22. L12:2208 excites
  23. L20:1749 suppresses
  24. L16:1973 excites

Activation histogram

How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.

1 values between -1.75 and -1.671 values between -1.32 and -1.231 values between -1.23 and -1.151 values between -1.15 and -1.064 values between -1.06 and -0.974 values between -0.97 and -0.895 values between -0.89 and -0.803 values between -0.80 and -0.7117 values between -0.71 and -0.6317 values between -0.63 and -0.5427 values between -0.54 and -0.4543 values between -0.45 and -0.3793 values between -0.37 and -0.28123 values between -0.28 and -0.19262 values between -0.19 and -0.11711 values between -0.11 and -0.021502 values between -0.02 and 0.07576 values between 0.07 and 0.15288 values between 0.15 and 0.24116 values between 0.24 and 0.3364 values between 0.33 and 0.4114 values between 0.41 and 0.505 values between 0.50 and 0.5913 values between 0.59 and 0.676 values between 0.67 and 0.762 values between 0.76 and 0.851 values between 0.85 and 0.933 values between 0.93 and 1.021 values between 1.02 and 1.111 values between 1.11 and 1.191 values between 1.37 and 1.451 values between 1.45 and 1.541 values between 1.63 and 1.71offnegativepositiveoften

Logit lens

The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layers instead.

Promotes

  1. anan
  2. __(*
  3. ibbon
  4. Throwable
  5. 影响
  6. 秀
  7. 受益
  8. Influence
  9. thích
  10. challenge

Suppresses

  1. ductor
  2. 밍
  3. нстру
  4. 拿来
  5. WXYZ
  6. erval
  7. 写道
  8. elsen
  9. errer
  10. ducted